On Wed, Apr 13, 2005 at 01:51:44PM +0000, Jason Stover wrote:
     
     > * test/expressions/randist.sh bails out on the beta and F distribution
     >   tests, but as per previous discussions, I think this is a problem
     >   with the test rather than the code.  I'll leave it to the numerical
     >   analysts to work out what's going on.
     
     I could not reproduce this problem with a recent checkout. I used a
     slightly modified version of randist.sh, as well as the usual
     randist.sh.  The modified randist.sh is pasted below. Running diff -b
     -B by hand returned no difference between the files written by
     randist.sh and the files in tests/expressions/randist/.
     

OK.  I dug  a bit deeper, and it seems that I had been testing against
an older version of tests/expressions/randist/{beta,f}.out (I don't
know why cvs update didn't do its job). 

But anyway, if one does:

cvs diff -r 1.1 -r 1.2 tests/expressions/randist/f.out

Then we get:

<      .99     5.00    10.00     5.64      .99      .01 
---
>      .99     5.00    10.00    10.00     1.00      .00 


I don't know why this change was made, but 10.00 is definitely wrong.
The correct answer is indeed 5.64

Similar changes to beta.out occured at the same time.  I don't know if
those changes are correct or not, because I don't have a book of
tables for the beta distribution.

I think I mentioned this before, but the idf of the F dist at 0.99
with 5;10 degrees of freedom is not an extreme case, and 10.00 when
the answer should be 5.6 is a very large error.

So it seems that an error has crept in, and we've fudged the test to
fit the broken results.
     

J'



-- 
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://pgp.mit.edu or any PGP keyserver for public key.


Attachment: pgpbylS9kPB6l.pgp
Description: PGP signature

_______________________________________________
pspp-dev mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/pspp-dev

Reply via email to